github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #7328] Performance degradation with 8B+ models on Windows Radeon #66711

New Issue
Closed
opened 2026-05-04 07:55:34 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-05-04 07:55:34 -05:00
Owner
Copy Link

Originally created by @7shi on GitHub (Oct 23, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/7328

Originally assigned to: @dhiltgen on GitHub.

What is the issue?

When running models 8B or larger on Windows with Radeon GPU, performance is slower than CPU-only mode, despite having sufficient VRAM available.

Environment:

  • OS: Windows 11 Home [10.0.22631]
  • CPU: AMD Ryzen 5 5600X 6-Core Processor
  • GPU: Radeon RX 7600 XT
  • VRAM: 16GB

Root Cause Investigation:

I've identified that this is caused by a HIP SDK behavior where memory allocations larger than 4GB are being redirected to shared GPU memory instead of using dedicated VRAM. I've reported this behavior to the HIP team here:

https://github.com/ROCm/HIP/issues/3644

Current Status:

As this is a HIP-level issue, improvement in model performance will depend on resolution from the HIP team. Creating this issue for visibility and to help others who might encounter similar performance degradation with large models on Windows Radeon setups.

Impact:

  • Models 8B and larger run slower than CPU-only mode
  • Available VRAM remains unused while slower shared memory is being utilized

379019648-866210ef-4a2c-4525-9026-9f614e19694e

OS

Windows

GPU

AMD

CPU

AMD

Ollama version

0.3.14

Originally created by @7shi on GitHub (Oct 23, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/7328 Originally assigned to: @dhiltgen on GitHub. ### What is the issue? When running models 8B or larger on Windows with Radeon GPU, performance is slower than CPU-only mode, despite having sufficient VRAM available. Environment: - OS: Windows 11 Home [10.0.22631] - CPU: AMD Ryzen 5 5600X 6-Core Processor - GPU: Radeon RX 7600 XT - VRAM: 16GB Root Cause Investigation: I've identified that this is caused by a HIP SDK behavior where memory allocations larger than 4GB are being redirected to shared GPU memory instead of using dedicated VRAM. I've reported this behavior to the HIP team here: https://github.com/ROCm/HIP/issues/3644 Current Status: As this is a HIP-level issue, improvement in model performance will depend on resolution from the HIP team. Creating this issue for visibility and to help others who might encounter similar performance degradation with large models on Windows Radeon setups. Impact: - Models 8B and larger run slower than CPU-only mode - Available VRAM remains unused while slower shared memory is being utilized ![379019648-866210ef-4a2c-4525-9026-9f614e19694e](https://github.com/user-attachments/assets/07a1739a-3ee0-4b68-b515-b5ee6c0f4b6f) ### OS Windows ### GPU AMD ### CPU AMD ### Ollama version 0.3.14
GiteaMirror added the bug label 2026-05-04 07:55:34 -05:00
GiteaMirror commented 2026-05-04 07:55:37 -05:00
Author
Owner
Copy Link

@0TTA commented on GitHub (Oct 23, 2024):

Same issue here!

<!-- gh-comment-id:2431752823 --> @0TTA commented on GitHub (Oct 23, 2024): Same issue here!
GiteaMirror commented 2026-05-04 07:55:38 -05:00
Author
Owner
Copy Link

@7shi commented on GitHub (Oct 23, 2024):

I've read your report - maybe the same issue.
https://github.com/ollama/ollama/issues/7330

<!-- gh-comment-id:2431909332 --> @7shi commented on GitHub (Oct 23, 2024): I've read your report - maybe the same issue. https://github.com/ollama/ollama/issues/7330
GiteaMirror commented 2026-05-04 07:55:38 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Oct 23, 2024):

This looks like a dup of #7107 - the only workaround for now is to downgrade to a driver OLDER than 24.9.1 until 24.11 comes out which should have the fix.

<!-- gh-comment-id:2432832082 --> @dhiltgen commented on GitHub (Oct 23, 2024): This looks like a dup of #7107 - the only workaround for now is to downgrade to a driver OLDER than 24.9.1 until 24.11 comes out which should have the fix.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#66711
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?